I can't get mapping ram work (Arm kernel 2.4 )

I can't get mapping ram work (Arm kernel 2.4 )

[ruben quiñones ruiz]

Hello everybody,

I'm developing a framebuffer display driver with an armv5 based architecture 
.
I've been able to send commands and display what I want using the Ioctl 
methods (the display works fine).

But now I want to add the framebuffer funcionality , so I wrote:

static int slcdc_mmap(struct file *filp, struct vm_area_struct *vma)
{
	unsigned long page,pos;
	unsigned long start = (unsigned long)vma->vm_start;
	unsigned long size = (unsigned long)(vma->vm_end-vma->vm_start);
			printk("\nMMAP CALLED");
	if(size > SLCDC_DATA_MEM_SIZE)
		return -EINVAL;
	pos = (unsigned long)slcdc_par.v_screen_start_address;
	while(size>0){
		page = virt_to_phys((void*)pos);
		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
		// This is an IO map - tell maydump to skip this VMA
		vma->vm_flags |= VM_IO;
		vma->vm_flags |= VM_RESERVED;
		if(remap_page_range(start,page,PAGE_SIZE,PAGE_SHARED)){
					printk("\n\nERRORRRR!!!!!!!!");
			return -EAGAIN;
		}
		start+= PAGE_SIZE;
		pos+= PAGE_SIZE;
		size-= PAGE_SIZE;
	}

	return 0;
}

Where slcdc_par.v_screen_start_address; is the virtual adress of the 
databuffer that must be sent (via a dma provided by the Cpu).

In user-space:

if((  pMapeo= (char *) mmap(NULL, screensize, PROT_READ | 
PROT_WRITE,MAP_SHARED,fichero,0))==NULL)  {
				 printf("Error al proyectar el archivo en memoria\n");
			 	return(4);
   			  }
			  if ((int)pMapeo == -1) {
       			 	printf("Error: failed to map device to memory.\n");
    			    exit(4);
  			  }
			  else
			  	printf("\ntoro Ok %x",pMapeo);
			 memset(pMapeo,0xffff,screensize);
			 for (aux=0;aux<screensize;aux++){
			 	printf("\n mapeo = %d",*(pMapeo+aux));
			 }
			 	aux=ioctl(fichero,SLCDC_CMD_RAMWR,&prueba);
I call to mmap and write successfully (printf shows all 0xff ...) and after 
that, I call the RAMWR ioctl:

case SLCDC_CMD_RAMWR:

			printk("RAMWR ");

			for (i=0;i<8;i++){
				for (j=0;j<127;j++){
					printk("\n Databuffer=%d",*(databuffer+i*128+j));
				}

				...send the data...
			}
		break;

Where databuffer is the physical address of 
slcdc_par.v_screen_start_address;

But, when I run it, printk shows that the memory is filled with 0's instead 
of the data written in user-space...

It seems like remap_page_range didn't work but I've been reading about all 
this process in google and I haven't found the solution to my problem... (if 
anyone wants to see how I initialize the buffer I'll post it... )

Yours faithfully,
Rubén Quiñones Ruiz -UAH University - Spain




-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: I can't get mapping ram work (Arm kernel 2.4 )

[Antonino A. Daplas]

ruben quiñones ruiz wrote:
> Hello everybody,
> 
> I'm developing a framebuffer display driver with an armv5 based 
> architecture .
> I've been able to send commands and display what I want using the Ioctl 
> methods (the display works fine).
> 
> But now I want to add the framebuffer funcionality , so I wrote:
> 
> static int slcdc_mmap(struct file *filp, struct vm_area_struct *vma)
> {
>     unsigned long page,pos;
>     unsigned long start = (unsigned long)vma->vm_start;
>     unsigned long size = (unsigned long)(vma->vm_end-vma->vm_start);
>             printk("\nMMAP CALLED");
>     if(size > SLCDC_DATA_MEM_SIZE)
>         return -EINVAL;
>     pos = (unsigned long)slcdc_par.v_screen_start_address;
>     while(size>0){
>         page = virt_to_phys((void*)pos);
>         vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
>         // This is an IO map - tell maydump to skip this VMA
>         vma->vm_flags |= VM_IO;
>         vma->vm_flags |= VM_RESERVED;
>         if(remap_page_range(start,page,PAGE_SIZE,PAGE_SHARED)){
>                     printk("\n\nERRORRRR!!!!!!!!");
>             return -EAGAIN;
>         }
>         start+= PAGE_SIZE;
>         pos+= PAGE_SIZE;
>         size-= PAGE_SIZE;
>     }
> 
>     return 0;
> }
> 
> Where slcdc_par.v_screen_start_address; is the virtual adress of the 
> databuffer that must be sent (via a dma provided by the Cpu).

How did you allocate this buffer?  Is is system RAM?  If it's system RAM,
did you use kmalloc, vmalloc or __get_free_pages. (kmalloc, vmalloc usually
do not work, you have to use __get_free_pages).

Tony


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: I can't get mapping ram work (Arm kernel 2.4 )

[ruben quiñones ruiz]

TONY wrote:
-----------
How did you allocate this buffer?  Is is system RAM?  If it's system RAM,
did you use kmalloc, vmalloc or __get_free_pages. (kmalloc, vmalloc usually
do not work, you have to use __get_free_pages).
  Tony
---------

Hi Tony, yeah , I use __get_free_pages : (this function is called in int 
init_module(void) )

int slcdc_init_buffer(void)
{
	u_int required_pages;
	u_int extra_pages;
	u_int order;
	struct page *page;
	u16  *allocated_region;

	printk("slcdc data buffer size = %x \n",(unsigned int)SLCDC_DATA_MEM_SIZE);

	if(g_slcdc_dbuffer_address != NULL)
		return -EINVAL;
	if(g_slcdc_cbuffer_address != NULL)
		return -EINVAL;
	/*find order required to allocate enough memory for it*/
	required_pages = SLCDC_DATA_MEM_SIZE >> PAGE_SHIFT;
	for (order = 0; required_pages >> order; order++){;}
	extra_pages = (1 << order) - required_pages;
	printk("PAGE_SHIFT=0x%x required_pages=0x%x\n",PAGE_SHIFT,required_pages);
	printk("extra_page=0x%x \n",extra_pages);

	if((allocated_region =	(u_char *)__get_free_pages(GFP_KERNEL | GFP_DMA, 
order)) == NULL)
	{
		printk("can not allocated memory\n");
       	return -ENOMEM;
	}

	g_slcdc_dbuffer_address = (u_char *)allocated_region + (extra_pages << 
PAGE_SHIFT);
	g_slcdc_dbuffer_phyaddress = (u_char 
*)__virt_to_phys((u_long)g_slcdc_dbuffer_address);
	printk("g_slcdc_dbuffer_address=0x%x \n",(unsigned 
int)g_slcdc_dbuffer_address);

    	/* Free all pages that we don't need but were given to us because */
   	 /* __get_free_pages() works on powers of 2. */
   	for (;extra_pages;extra_pages--)
		free_page((u_int)allocated_region + ((extra_pages-1) << PAGE_SHIFT));

	/* Set reserved flag for fb memory to allow it to be remapped into */
	/* user space by the common fbmem driver using remap_page_range(). */
	for(page = virt_to_page(g_slcdc_dbuffer_address);
		page < virt_to_page(g_slcdc_dbuffer_address + SLCDC_DATA_MEM_SIZE);
	    page++)
	{
		mem_map_reserve(page);
	}
	slcdc_par.screen_start_address  
=(u_char*)((u_long)g_slcdc_dbuffer_phyaddress);
	slcdc_par.v_screen_start_address 
=(u_char*)((u_long)g_slcdc_dbuffer_address);
	printk("startaddr= 0x%x, phyaddr=0x%x , size=0x%x \n",(unsigned 
int)slcdc_par.screen_start_address,(unsigned 
int)slcdc_par.v_screen_start_address,(unsigned int)SLCDC_DATA_MEM_SIZE);
	memset(slcdc_par.v_screen_start_address,0,SLCDC_DATA_MEM_SIZE);

	g_slcdc_cbuffer_address = 
consistent_alloc(GFP_KERNEL|GFP_DMA,(SLCDC_CMD_MEM_SIZE+4),&g_slcdc_cbuffer_phyaddress);
	g_slcdc_cbuffer_address = (u_long)g_slcdc_cbuffer_address & 0xfffffff4;
	g_slcdc_cbuffer_phyaddress = (u_long)g_slcdc_cbuffer_phyaddress & 
0xfffffff4;

	printk("slcdc data buffer address = %x cmd buffer address= %x \n",(unsigned 
int)slcdc_par.screen_start_address,(unsigned int) g_slcdc_cbuffer_address);

	return 0;
}

I think , I must have a stupid error that fuck's me.. but I don't find it 
:-p

Yours Faithfully,
Rubén Quiñones Ruiz -- UAH University -- Spain




-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: I can't get mapping ram work (Arm kernel 2.4 )

[Antonino A. Daplas]

ruben quiñones ruiz wrote:
> TONY wrote:
> -----------
> How did you allocate this buffer?  Is is system RAM?  If it's system RAM,
> did you use kmalloc, vmalloc or __get_free_pages. (kmalloc, vmalloc usually
> do not work, you have to use __get_free_pages).
>  Tony
> ---------
> 
> Hi Tony, yeah , I use __get_free_pages : (this function is called in int 
> init_module(void) )
> 

Hmm, I don't see anything obviously wrong with the code.  Maybe you can
post this at the ARM mailing list (linux-arm-kernel@lists.arm.linux.org.uk).

Tony


-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf

Re: I can't get mapping ram work (Arm kernel 2.4 )

[ruben quiñones ruiz]

Tony wrote:

Hmm, I don't see anything obviously wrong with the code.  Maybe you can
post this at the ARM mailing list (linux-arm-kernel@lists.arm.linux.org.uk).
Tony


Thanks for the answer... I'm suscribing in the Arm mailing list :-)
On the other hand, I have another GPL driver for another display using the 
same code that works perfectly.. so I thought it was a stupid mistake...

Best Regards,
Rubén Quiñones Ruiz  --UAH University -- Spain




-------------------------------------------------------
SF.Net email is Sponsored by the Better Software Conference & EXPO
September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf